1Using RCU's CPU Stall Detector
2
3The rcu_cpu_stall_suppress module parameter enables RCU's CPU stall
4detector, which detects conditions that unduly delay RCU grace periods.
5This module parameter enables CPU stall detection by default, but
6may be overridden via boot-time parameter or at runtime via sysfs.
7The stall detector's idea of what constitutes "unduly delayed" is
8controlled by a set of kernel configuration variables and cpp macros:
9
10CONFIG_RCU_CPU_STALL_TIMEOUT
11
12	This kernel configuration parameter defines the period of time
13	that RCU will wait from the beginning of a grace period until it
14	issues an RCU CPU stall warning.  This time period is normally
15	21 seconds.
16
17	This configuration parameter may be changed at runtime via the
18	/sys/module/rcupdate/parameters/rcu_cpu_stall_timeout, however
19	this parameter is checked only at the beginning of a cycle.
20	So if you are 10 seconds into a 40-second stall, setting this
21	sysfs parameter to (say) five will shorten the timeout for the
22	-next- stall, or the following warning for the current stall
23	(assuming the stall lasts long enough).  It will not affect the
24	timing of the next warning for the current stall.
25
26	Stall-warning messages may be enabled and disabled completely via
27	/sys/module/rcupdate/parameters/rcu_cpu_stall_suppress.
28
29CONFIG_RCU_CPU_STALL_INFO
30
31	This kernel configuration parameter causes the stall warning to
32	print out additional per-CPU diagnostic information, including
33	information on scheduling-clock ticks and RCU's idle-CPU tracking.
34
35RCU_STALL_DELAY_DELTA
36
37	Although the lockdep facility is extremely useful, it does add
38	some overhead.  Therefore, under CONFIG_PROVE_RCU, the
39	RCU_STALL_DELAY_DELTA macro allows five extra seconds before
40	giving an RCU CPU stall warning message.  (This is a cpp
41	macro, not a kernel configuration parameter.)
42
43RCU_STALL_RAT_DELAY
44
45	The CPU stall detector tries to make the offending CPU print its
46	own warnings, as this often gives better-quality stack traces.
47	However, if the offending CPU does not detect its own stall in
48	the number of jiffies specified by RCU_STALL_RAT_DELAY, then
49	some other CPU will complain.  This delay is normally set to
50	two jiffies.  (This is a cpp macro, not a kernel configuration
51	parameter.)
52
53rcupdate.rcu_task_stall_timeout
54
55	This boot/sysfs parameter controls the RCU-tasks stall warning
56	interval.  A value of zero or less suppresses RCU-tasks stall
57	warnings.  A positive value sets the stall-warning interval
58	in jiffies.  An RCU-tasks stall warning starts wtih the line:
59
60		INFO: rcu_tasks detected stalls on tasks:
61
62	And continues with the output of sched_show_task() for each
63	task stalling the current RCU-tasks grace period.
64
65For non-RCU-tasks flavors of RCU, when a CPU detects that it is stalling,
66it will print a message similar to the following:
67
68INFO: rcu_sched_state detected stall on CPU 5 (t=2500 jiffies)
69
70This message indicates that CPU 5 detected that it was causing a stall,
71and that the stall was affecting RCU-sched.  This message will normally be
72followed by a stack dump of the offending CPU.  On TREE_RCU kernel builds,
73RCU and RCU-sched are implemented by the same underlying mechanism,
74while on PREEMPT_RCU kernel builds, RCU is instead implemented
75by rcu_preempt_state.
76
77On the other hand, if the offending CPU fails to print out a stall-warning
78message quickly enough, some other CPU will print a message similar to
79the following:
80
81INFO: rcu_bh_state detected stalls on CPUs/tasks: { 3 5 } (detected by 2, 2502 jiffies)
82
83This message indicates that CPU 2 detected that CPUs 3 and 5 were both
84causing stalls, and that the stall was affecting RCU-bh.  This message
85will normally be followed by stack dumps for each CPU.  Please note that
86PREEMPT_RCU builds can be stalled by tasks as well as by CPUs,
87and that the tasks will be indicated by PID, for example, "P3421".
88It is even possible for a rcu_preempt_state stall to be caused by both
89CPUs -and- tasks, in which case the offending CPUs and tasks will all
90be called out in the list.
91
92Finally, if the grace period ends just as the stall warning starts
93printing, there will be a spurious stall-warning message:
94
95INFO: rcu_bh_state detected stalls on CPUs/tasks: { } (detected by 4, 2502 jiffies)
96
97This is rare, but does happen from time to time in real life.  It is also
98possible for a zero-jiffy stall to be flagged in this case, depending
99on how the stall warning and the grace-period initialization happen to
100interact.  Please note that it is not possible to entirely eliminate this
101sort of false positive without resorting to things like stop_machine(),
102which is overkill for this sort of problem.
103
104If the CONFIG_RCU_CPU_STALL_INFO kernel configuration parameter is set,
105more information is printed with the stall-warning message, for example:
106
107	INFO: rcu_preempt detected stall on CPU
108	0: (63959 ticks this GP) idle=241/3fffffffffffffff/0 softirq=82/543
109	   (t=65000 jiffies)
110
111In kernels with CONFIG_RCU_FAST_NO_HZ, even more information is
112printed:
113
114	INFO: rcu_preempt detected stall on CPU
115	0: (64628 ticks this GP) idle=dd5/3fffffffffffffff/0 softirq=82/543 last_accelerate: a345/d342 nonlazy_posted: 25 .D
116	   (t=65000 jiffies)
117
118The "(64628 ticks this GP)" indicates that this CPU has taken more
119than 64,000 scheduling-clock interrupts during the current stalled
120grace period.  If the CPU was not yet aware of the current grace
121period (for example, if it was offline), then this part of the message
122indicates how many grace periods behind the CPU is.
123
124The "idle=" portion of the message prints the dyntick-idle state.
125The hex number before the first "/" is the low-order 12 bits of the
126dynticks counter, which will have an even-numbered value if the CPU is
127in dyntick-idle mode and an odd-numbered value otherwise.  The hex
128number between the two "/"s is the value of the nesting, which will
129be a small positive number if in the idle loop and a very large positive
130number (as shown above) otherwise.
131
132The "softirq=" portion of the message tracks the number of RCU softirq
133handlers that the stalled CPU has executed.  The number before the "/"
134is the number that had executed since boot at the time that this CPU
135last noted the beginning of a grace period, which might be the current
136(stalled) grace period, or it might be some earlier grace period (for
137example, if the CPU might have been in dyntick-idle mode for an extended
138time period.  The number after the "/" is the number that have executed
139since boot until the current time.  If this latter number stays constant
140across repeated stall-warning messages, it is possible that RCU's softirq
141handlers are no longer able to execute on this CPU.  This can happen if
142the stalled CPU is spinning with interrupts are disabled, or, in -rt
143kernels, if a high-priority process is starving RCU's softirq handler.
144
145For CONFIG_RCU_FAST_NO_HZ kernels, the "last_accelerate:" prints the
146low-order 16 bits (in hex) of the jiffies counter when this CPU last
147invoked rcu_try_advance_all_cbs() from rcu_needs_cpu() or last invoked
148rcu_accelerate_cbs() from rcu_prepare_for_idle().  The "nonlazy_posted:"
149prints the number of non-lazy callbacks posted since the last call to
150rcu_needs_cpu().  Finally, an "L" indicates that there are currently
151no non-lazy callbacks ("." is printed otherwise, as shown above) and
152"D" indicates that dyntick-idle processing is enabled ("." is printed
153otherwise, for example, if disabled via the "nohz=" kernel boot parameter).
154
155If the relevant grace-period kthread has been unable to run prior to
156the stall warning, the following additional line is printed:
157
158	rcu_preempt kthread starved for 2023 jiffies!
159
160Starving the grace-period kthreads of CPU time can of course result in
161RCU CPU stall warnings even when all CPUs and tasks have passed through
162the required quiescent states.
163
164
165Multiple Warnings From One Stall
166
167If a stall lasts long enough, multiple stall-warning messages will be
168printed for it.  The second and subsequent messages are printed at
169longer intervals, so that the time between (say) the first and second
170message will be about three times the interval between the beginning
171of the stall and the first message.
172
173
174What Causes RCU CPU Stall Warnings?
175
176So your kernel printed an RCU CPU stall warning.  The next question is
177"What caused it?"  The following problems can result in RCU CPU stall
178warnings:
179
180o	A CPU looping in an RCU read-side critical section.
181	
182o	A CPU looping with interrupts disabled.  This condition can
183	result in RCU-sched and RCU-bh stalls.
184
185o	A CPU looping with preemption disabled.  This condition can
186	result in RCU-sched stalls and, if ksoftirqd is in use, RCU-bh
187	stalls.
188
189o	A CPU looping with bottom halves disabled.  This condition can
190	result in RCU-sched and RCU-bh stalls.
191
192o	For !CONFIG_PREEMPT kernels, a CPU looping anywhere in the
193	kernel without invoking schedule().  Note that cond_resched()
194	does not necessarily prevent RCU CPU stall warnings.  Therefore,
195	if the looping in the kernel is really expected and desirable
196	behavior, you might need to replace some of the cond_resched()
197	calls with calls to cond_resched_rcu_qs().
198
199o	Anything that prevents RCU's grace-period kthreads from running.
200	This can result in the "All QSes seen" console-log message.
201	This message will include information on when the kthread last
202	ran and how often it should be expected to run.
203
204o	A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might
205	happen to preempt a low-priority task in the middle of an RCU
206	read-side critical section.   This is especially damaging if
207	that low-priority task is not permitted to run on any other CPU,
208	in which case the next RCU grace period can never complete, which
209	will eventually cause the system to run out of memory and hang.
210	While the system is in the process of running itself out of
211	memory, you might see stall-warning messages.
212
213o	A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that
214	is running at a higher priority than the RCU softirq threads.
215	This will prevent RCU callbacks from ever being invoked,
216	and in a CONFIG_PREEMPT_RCU kernel will further prevent
217	RCU grace periods from ever completing.  Either way, the
218	system will eventually run out of memory and hang.  In the
219	CONFIG_PREEMPT_RCU case, you might see stall-warning
220	messages.
221
222o	A hardware or software issue shuts off the scheduler-clock
223	interrupt on a CPU that is not in dyntick-idle mode.  This
224	problem really has happened, and seems to be most likely to
225	result in RCU CPU stall warnings for CONFIG_NO_HZ_COMMON=n kernels.
226
227o	A bug in the RCU implementation.
228
229o	A hardware failure.  This is quite unlikely, but has occurred
230	at least once in real life.  A CPU failed in a running system,
231	becoming unresponsive, but not causing an immediate crash.
232	This resulted in a series of RCU CPU stall warnings, eventually
233	leading the realization that the CPU had failed.
234
235The RCU, RCU-sched, RCU-bh, and RCU-tasks implementations have CPU stall
236warning.  Note that SRCU does -not- have CPU stall warnings.  Please note
237that RCU only detects CPU stalls when there is a grace period in progress.
238No grace period, no CPU stall warnings.
239
240To diagnose the cause of the stall, inspect the stack traces.
241The offending function will usually be near the top of the stack.
242If you have a series of stall warnings from a single extended stall,
243comparing the stack traces can often help determine where the stall
244is occurring, which will usually be in the function nearest the top of
245that portion of the stack which remains the same from trace to trace.
246If you can reliably trigger the stall, ftrace can be quite helpful.
247
248RCU bugs can often be debugged with the help of CONFIG_RCU_TRACE
249and with RCU's event tracing.  For information on RCU's event tracing,
250see include/trace/events/rcu.h.
251